* [PATCH] balance_dirty_pages - exit loop when no more pages available
@ 2007-07-31 17:11 richard kennedy
2007-08-02 19:54 ` Chuck Ebbert
0 siblings, 1 reply; 2+ messages in thread
From: richard kennedy @ 2007-07-31 17:11 UTC (permalink / raw)
To: linux-mm; +Cc: linux-kernel
exit loop in balance_dirty_pages when no more pages available to write
On a bdi that has very little traffic balance_dirty_pages can loop
needlessly waiting until do_writepages has written enough pages.
do_writepages will return encountered_congestion==0 && nr_to_write > 0
when it has completed a pass but did not find enough pages available to
write. balance_dirty_pages ignores this and keeps looping until a total
of chunk pages was written.
this patch adds an extra exit condition to break out of the loop in this
case.
I've tested this on my amd64 desktop, and I also have a version of this
patch that includes a printk and a test case that occasionally does
trigger this condition.
Signed-off-by: Richard Kennedy <richard@rsk.demon.co.uk>
------
--- linux-2.6.22.1/mm/page-writeback.c.orig 2007-07-30 16:36:09.000000000 +0100
+++ linux-2.6.22.1/mm/page-writeback.c 2007-07-31 16:26:43.000000000 +0100
@@ -250,6 +250,8 @@ static void balance_dirty_pages(struct a
pages_written += write_chunk - wbc.nr_to_write;
if (pages_written >= write_chunk)
break; /* We've done our duty */
+ if (!wbc.encountered_congestion && wbc.nr_to_write > 0)
+ break; /* didn't find enough to do */
}
congestion_wait(WRITE, HZ/10);
}
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH] balance_dirty_pages - exit loop when no more pages available
2007-07-31 17:11 [PATCH] balance_dirty_pages - exit loop when no more pages available richard kennedy
@ 2007-08-02 19:54 ` Chuck Ebbert
0 siblings, 0 replies; 2+ messages in thread
From: Chuck Ebbert @ 2007-08-02 19:54 UTC (permalink / raw)
To: Peter Zijlstra; +Cc: richard kennedy, linux-mm, linux-kernel
On 07/31/2007 01:11 PM, richard kennedy wrote:
Peter, did you see this?
> exit loop in balance_dirty_pages when no more pages available to write
>
> On a bdi that has very little traffic balance_dirty_pages can loop
> needlessly waiting until do_writepages has written enough pages.
>
> do_writepages will return encountered_congestion==0 && nr_to_write > 0
> when it has completed a pass but did not find enough pages available to
> write. balance_dirty_pages ignores this and keeps looping until a total
> of chunk pages was written.
>
> this patch adds an extra exit condition to break out of the loop in this
> case.
>
> I've tested this on my amd64 desktop, and I also have a version of this
> patch that includes a printk and a test case that occasionally does
> trigger this condition.
>
> Signed-off-by: Richard Kennedy <richard@rsk.demon.co.uk>
>
> ------
> --- linux-2.6.22.1/mm/page-writeback.c.orig 2007-07-30 16:36:09.000000000 +0100
> +++ linux-2.6.22.1/mm/page-writeback.c 2007-07-31 16:26:43.000000000 +0100
> @@ -250,6 +250,8 @@ static void balance_dirty_pages(struct a
> pages_written += write_chunk - wbc.nr_to_write;
> if (pages_written >= write_chunk)
> break; /* We've done our duty */
> + if (!wbc.encountered_congestion && wbc.nr_to_write > 0)
> + break; /* didn't find enough to do */
> }
> congestion_wait(WRITE, HZ/10);
> }
>
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2007-08-02 19:54 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2007-07-31 17:11 [PATCH] balance_dirty_pages - exit loop when no more pages available richard kennedy
2007-08-02 19:54 ` Chuck Ebbert
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox